Carlo Curino commented on YARN-2009:

[~sunilg], pardon the long delay... From what you say it seems like the 
priority issues within queue it is important for you and you observe 
non-trivial delays. 
If that's is the case, I think it is fine to venture in adding within-queue 
cross-app preemption. I would argue in favor of a conservative policy with 
several built-in
dampers (like max per-round preemptions, deadzones, fraction of imbalance) like 
we did for cross-queue. Also we should be careful not to make it too expensive
(if we have thousands of apps in the queues, we should be mindful of not 
overloading the RM with costly rebalancing algos, and extra scheduling 
decisions that
derived from preemption).

What [~eepayne] says also makes sense, if we preemptions triggered by 
cross-queue imbalances it would be good to "spend" them to correct the issue 
you observed.

> Priority support for preemption in ProportionalCapacityPreemptionPolicy
> -----------------------------------------------------------------------
>                 Key: YARN-2009
>                 URL: https://issues.apache.org/jira/browse/YARN-2009
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>          Components: capacityscheduler
>            Reporter: Devaraj K
>            Assignee: Sunil G
> While preempting containers based on the queue ideal assignment, we may need 
> to consider preempting the low priority application containers first.

This message was sent by Atlassian JIRA

Reply via email to